Genome-wide association and optimization of prediction accuracies on genomic selection models for Eucalyptus grandis W. Hill

• Ano de defesa: 2022
• Autor(a) principal: Rocha, Lucas Fernandes [UNESP]
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: genome-wide association; genomic selection; multi-trait analysis; forest breeding; eucalypt; associação genômica ampla; seleção genômica; análise multi-trait; melhoramento florestal; eucalipto
• Link de acesso: http://hdl.handle.net/11449/234813

Molecular markers that are widely distributed throughout the genome offer a fundamental tool to optimize forest tree breeding programs. This study aimed to evaluate the genetic architecture of quantitative genes and optimize genomic selection models for growth and wood-quality traits of Eucalyptus grandis. We evaluated an open-pollinated breeding population with 1,772 genotypes, composed of 27 different families, that was established using complete randomized block design with 20 replicates each. Individuals were genotyped using the Illumina Infinium EuCHIP60K chip and 12 different phenotypic variables were evaluated for growth traits (diameter at breast height, height, and volume evaluated at 3 and 6 years after planting) and wood-quality traits (pure cellulose yield, basic wood density, syringyl/guayacil, soluble lignin, total solids, and total extractives). First, we performed a genome-wide association study (GWAS) using the single-trait model (farmCPU) and multi-trait (MTMM) mixed models. Next, we searched for quantitative trait loci (QTLs) and their predicted functional effects using a database for Eucalyptus. Subsequently, the accuracy of the prediction ability, coincidence of selection, and selection gains of genomic selection models were analyzed based on the Genomic Best Linear Unbiased Prediction (GBLUP) method. We tested different approaches considering the additive variance, additive-dominant variance, optimization of training set, and multi-trait models. Finally, we analyzed the efficiency of using growth traits to increase the prediction ability of wood-quality traits considering a multi-trait model with optimization of training set methodology. After quality control, a total of 21,254 informative SNPs were found that have a wide distribution and a high linkage disequilibrium decay across the 11 chromosomes. For the GWAS analysis, the farmCPU model identified 43 and 38 small effect markers that are significantly associated with growth and wood quality traits, respectively. Similarly, pleiotropic SNPs were also discovered between growth (24) and wood quality traits (6) using the MTMM model. Through gene ontology analysis, we identified genes responsible for plant growth and related with hydric stress. For the genomic selection analysis, growth traits appeared to be more influenced by dominance than wood quality traits, meanwhile GBLUP models were effective in predicting wood quality traits. Although the results for CS appear to be low, SG values were relatively high. The optimization of the training set analysis effectively selected the best genotypes to be used as the training set. Additionally, the multi-trait and multi-trait with optimization of the training set were also able to increase the prediction ability of the GBLUP models. Thus, information from growth traits can be used to effectively increase the prediction ability of wood quality traits. Our study demonstrates the complex nature of quantitative traits, provides new evidence for the architecture of genes related to trait expression, and highlights the efficiency of genomic selection models to predict phenotypic expression in E. grandis.